TimeWorld improves forecasting by modeling action effects on time series
TimeWorld: Towards Action-Conditioned Multimodal World Models for Time Series Forecasting
Machine Learning
Summary
Predicting what will happen next is hard when actions change the future, like sales promotions affecting retail demand. The authors found that existing models struggle to account for actions described in text when forecasting time-dependent events. They made a new test set and showed that adapting large pretrained models to better understand these actions improves predictions. They then created TimeWorld, a method that links actions to how they change future events, helping models forecast more accurately when actions occur.
What this means in practice
- •For online retail teams: Improve demand forecasting by modeling how planned marketing actions shape future sales trends.
- •For network operation centers: Enhance network usage and load forecasts by incorporating planned interventions or configuration changes as text inputs.
Authors
Hang Ni, Weijia Zhang, Hao Liu
Abstract
Time series forecasting supports decision-making in web applications such as online retail and network management, where planned actions can reshape future dynamics. World models provide a natural paradigm for predicting such action-dependent changes beyond extrapolating historical patterns. However, existing time series world models (TSWMs) typically rely on fixed, domain-specific covariate schemas that limit flexible intervention specification, while multimodal forecasters treat text primarily as passive predictive context. To investigate these gaps, we introduce TSWM-Bench, a cross-domain benchmark for multimodal TSWMs spanning seven real-world datasets, with textual interventions grounded in recorded action--outcome data. Our evaluation reveals that conventional multimodal forecasters underperform history-only time series foundation models (TSFMs) despite dataset-specific training, whereas simple text-prefix adaptations improve forecasting across all evaluated TSFM backbones. These findings suggest that pretrained TSFMs encode implicit world knowledge of temporal dynamics, offering a strong foundation for action-conditioned modeling beyond conventional multimodal fusion. Building on this insight, we propose TimeWorld, a plug-in framework that adapts pretrained TSFMs through explicit action-response modeling. Specifically, the Prototype-Guided Latent Modulation (PLM) module grounds textual actions in temporal dynamics by composing representative response prototypes to modulate future representations based on the action and historical patterns. The Action Contrastive Learning (ACL) component strengthens action-specific response modeling by contrasting these updates under alternative interventions sharing the same history, distinguishing action types while preserving relative intensity relationships.